1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
|
/*========================== begin_copyright_notice ============================
Copyright (C) 2017-2021 Intel Corporation
SPDX-License-Identifier: MIT
============================= end_copyright_notice ===========================*/
#include "Compiler/SampleCmpToDiscard.h"
#include "Compiler/CodeGenPublic.h"
#include "Compiler/IGCPassSupport.h"
#include "GenISAIntrinsics/GenIntrinsicInst.h"
using namespace llvm;
using namespace IGC;
using namespace GenISAIntrinsic;
class SampleCmpToDiscard : public FunctionPass
{
public:
static char ID;
SampleCmpToDiscard() : FunctionPass(ID)
{
}
virtual void getAnalysisUsage(llvm::AnalysisUsage& AU) const
{
AU.addRequired<CodeGenContextWrapper>();
}
virtual bool runOnFunction(Function& F);
virtual llvm::StringRef getPassName() const
{
return "SampleCmpToDiscard";
}
private:
bool processBlock(BasicBlock* BB);
bool canFoldValue(Instruction* inst, std::map<Value*, APFloat>& instToValMap);
};
char SampleCmpToDiscard::ID = 0;
#define PASS_FLAG "igc-discard-samplecmp"
#define PASS_DESCRIPTION "Checks possibility of SampleCmpToDiscard optimization"
#define PASS_CFG_ONLY false
#define PASS_ANALYSIS false
IGC_INITIALIZE_PASS_BEGIN(SampleCmpToDiscard, PASS_FLAG, PASS_DESCRIPTION, PASS_CFG_ONLY, PASS_ANALYSIS)
IGC_INITIALIZE_PASS_END(SampleCmpToDiscard, PASS_FLAG, PASS_DESCRIPTION, PASS_CFG_ONLY, PASS_ANALYSIS)
FunctionPass* IGC::CreateSampleCmpToDiscardPass()
{
return new SampleCmpToDiscard();
}
bool SampleCmpToDiscard::runOnFunction(Function& F)
{
auto pCtx = getAnalysis<CodeGenContextWrapper>().getCodeGenContext();
if (pCtx->type != ShaderType::PIXEL_SHADER)
{
return false;
}
bool changed = false;
for (auto BI = F.begin(), BE = F.end(); BI != BE;)
{
BasicBlock* currentBB = &(*BI);
++BI;
changed |= processBlock(currentBB);
}
return changed;
}
bool SampleCmpToDiscard::processBlock(BasicBlock* BB)
{
auto pCtx = getAnalysis<CodeGenContextWrapper>().getCodeGenContext();
std::map<Value*, APFloat> instToValMap;
for (auto II = BB->rbegin(), IE = BB->rend(); II != IE; ++II)
{
instToValMap.clear();
if (GenIntrinsicInst * genIntr = dyn_cast<GenIntrinsicInst>(&(*II)))
{
Instruction* valueToFold = nullptr;
GenISAIntrinsic::ID ID = genIntr->getIntrinsicID();
// only consider sample_lc
if (ID == GenISAIntrinsic::GenISA_sampleCptr ||
ID == GenISAIntrinsic::GenISA_sampleBCptr)
{
if (genIntr->hasOneUse())
{
if (ExtractElementInst * ext = dyn_cast<ExtractElementInst>(*genIntr->user_begin()))
{
if (ConstantInt * index = dyn_cast<ConstantInt>(ext->getIndexOperand()))
{
if (index->isZero())
{
valueToFold = ext;
}
}
}
}
}
// if we haven't found the pattern we want to skip to the next instruction
if (valueToFold == nullptr)
{
continue;
}
APFloat newAPFloat = cast<ConstantFP>(ConstantFP::get(valueToFold->getType(), 1.0))->getValueAPF();
instToValMap.emplace(valueToFold, newAPFloat);
bool foldableInst = canFoldValue(valueToFold, instToValMap);
if (foldableInst)
{
llvm::Value* texOp = cast<SampleIntrinsic>(genIntr)->getTextureValue();
// set compiler outputs
if (texOp->getType()->getPointerAddressSpace())
{
uint as = texOp->getType()->getPointerAddressSpace();
uint bufferIndex;
bool directIndexing;
DecodeAS4GFXResource(as, directIndexing, bufferIndex);
// set the compiler outputs to indicate sample_Cqa which could write 1.0f to the driver
pCtx->m_instrTypes.sampleCmpToDiscardOptimizationPossible = true;
pCtx->m_instrTypes.sampleCmpToDiscardOptimizationSlot = bufferIndex;
return true;
}
}
}
}
instToValMap.clear();
return false;
}
bool SampleCmpToDiscard::canFoldValue(Instruction* inst, std::map<Value*, APFloat>& instToValMap)
{
// currently we are handling the following type of cases
/*
* %92 = call <4 x half> @llvm.genx.GenISA.sampleCptr.v4f16.f16.p196609v4f32.p524289i32(
half %91, half %89, half %90, half %91, half %87, half 0xH0000, <4 x float> addrspace(196609)* null,
i32 addrspace(524289)* null, i32 0, i32 0, i32 0)
* %93 = extractelement <4 x half> %92, i32 0
* %94 = fsub half 0xH3C00, %93 // 1 - B
* %95 = fmul half %94, 0xH3400 // 0.25 * (1-B)
* %96 = fadd half %95, %93 // 0.25 * ( 1-B) + B
* call void @llvm.genx.GenISA.OUTPUT.f16(half %96, half %96, half %96, half %96, i32 0, i32 0)
* ret void
*/
/*
* %62 = call <4 x float> @llvm.genx.GenISA.sampleBCptr.v4f32.f32.p196609i8.p524293i8(
* float %61, float 0.000000e+00, float %57, float %58, float 0.000000e+00,
* float 0.000000e+00, i8 addrspace(196609)* null,
* i8 addrspace(524293)* inttoptr (i64 5 to i8 addrspace(524293)*), i32 0, i32 0, i32 0)
* %scalar40 = extractelement <4 x float> %62, i32 0
* %phitmp = fmul float %scalar40, 7.500000e-01
* %63 = fadd float %phitmp, 2.500000e-01
* call void @llvm.genx.GenISA.OUTPUT.f32(float %63, float %63, float %63, float %63, i32 0, i32 0)
*/
bool foldInst = false;
APFloat newConstantFloat = cast<ConstantFP>(ConstantFP::get(inst->getType(), 1.0))->getValueAPF();
auto mapFind = instToValMap.find(inst);
if (mapFind == instToValMap.end())
{
return false;
}
for (auto UI = inst->user_begin(), UE = inst->user_end(); UI != UE; ++UI)
{
// assume the value of inst returns value 1;
BinaryOperator* bin = dyn_cast<BinaryOperator>(*UI);
if (bin && bin->getOpcode() == Instruction::FSub)
{
auto itOtherOp = (bin->getOperand(0) == inst) ? instToValMap.find(bin->getOperand(1)) : instToValMap.find(bin->getOperand(0));
if (isa<ConstantFP>(bin->getOperand(0)))
{
ConstantFP* C0 = dyn_cast<ConstantFP>(bin->getOperand(0));
newConstantFloat = C0->getValueAPF();
newConstantFloat.subtract(mapFind->second, llvm::APFloat::rmNearestTiesToEven);
instToValMap.emplace(bin, newConstantFloat);
foldInst = canFoldValue(bin, instToValMap);
}
else if (isa<ConstantFP>(bin->getOperand(1)))
{
ConstantFP* C0 = dyn_cast<ConstantFP>(bin->getOperand(1));
newConstantFloat = mapFind->second;
newConstantFloat.subtract(C0->getValueAPF(), llvm::APFloat::rmNearestTiesToEven);
instToValMap.emplace(bin, newConstantFloat);
foldInst = canFoldValue(bin, instToValMap);
}
else if (itOtherOp != instToValMap.end())
{
if (bin->getOperand(0) == inst)
{
newConstantFloat = mapFind->second;
newConstantFloat.subtract(itOtherOp->second, llvm::APFloat::rmNearestTiesToEven);
}
else
{
newConstantFloat = itOtherOp->second;
newConstantFloat.subtract(mapFind->second, llvm::APFloat::rmNearestTiesToEven);
}
instToValMap.emplace(bin, newConstantFloat);
foldInst = canFoldValue(bin, instToValMap);
}
}
else if (bin && bin->getOpcode() == Instruction::FAdd)
{
auto itOtherOp = (bin->getOperand(0) == inst) ? instToValMap.find(bin->getOperand(1)) : instToValMap.find(bin->getOperand(0));
if (isa<ConstantFP>(bin->getOperand(0)))
{
ConstantFP* C0 = dyn_cast<ConstantFP>(bin->getOperand(0));
newConstantFloat = mapFind->second;
newConstantFloat.add(C0->getValueAPF(), llvm::APFloat::rmNearestTiesToEven);
instToValMap.emplace(bin, newConstantFloat);
foldInst = canFoldValue(bin, instToValMap);
}
else if (isa<ConstantFP>(bin->getOperand(1)))
{
ConstantFP* C0 = dyn_cast<ConstantFP>(bin->getOperand(1));
newConstantFloat = mapFind->second;
newConstantFloat.add(C0->getValueAPF(), llvm::APFloat::rmNearestTiesToEven);
instToValMap.emplace(bin, newConstantFloat);
foldInst = canFoldValue(bin, instToValMap);
}
else if (itOtherOp != instToValMap.end())
{
newConstantFloat = mapFind->second;
newConstantFloat.add(itOtherOp->second, llvm::APFloat::rmNearestTiesToEven);
instToValMap.emplace(bin, newConstantFloat);
foldInst = canFoldValue(bin, instToValMap);
}
}
else if (bin && bin->getOpcode() == Instruction::FMul)
{
auto itOtherOp = (bin->getOperand(0) == inst) ? instToValMap.find(bin->getOperand(1)) : instToValMap.find(bin->getOperand(0));
if (isa<ConstantFP>(bin->getOperand(0)))
{
ConstantFP* C0 = dyn_cast<ConstantFP>(bin->getOperand(0));
newConstantFloat = mapFind->second;
newConstantFloat.multiply(C0->getValueAPF(), llvm::APFloat::rmNearestTiesToEven);
instToValMap.emplace(bin, newConstantFloat);
foldInst = canFoldValue(bin, instToValMap);
}
else if (isa<ConstantFP>(bin->getOperand(1)))
{
ConstantFP* C0 = dyn_cast<ConstantFP>(bin->getOperand(1));
newConstantFloat = mapFind->second;
newConstantFloat.multiply(C0->getValueAPF(), llvm::APFloat::rmNearestTiesToEven);
instToValMap.emplace(bin, newConstantFloat);
foldInst = canFoldValue(bin, instToValMap);
}
else if (itOtherOp != instToValMap.end())
{
newConstantFloat = mapFind->second;
newConstantFloat.multiply(itOtherOp->second, llvm::APFloat::rmNearestTiesToEven);
instToValMap.emplace(bin, newConstantFloat);
foldInst = canFoldValue(bin, instToValMap);
}
}
else if (isa<GenIntrinsicInst>(*UI) && ((mapFind->second).compare(newConstantFloat) == APFloat::cmpEqual))
{
GenIntrinsicInst* CI = dyn_cast<GenIntrinsicInst>(*UI);
if ((CI->getIntrinsicID() == GenISA_RTWrite) && (llvm::isa<llvm::ReturnInst>(CI->getNextNode())))
{
return true;
}
}
}
return foldInst;
}
|
